Articles: Memory
 

Bookmark and Share

(9) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 ]

Performance

In our performance tests we checked out the speed of our LGA1155 platform with different dual-channel 32 GB memory kits from G.Skill. Each kit was tested in two modes: 1) automatic configuring when every timing is set by the mainboard's BIOS according to the XMP data or official specifications (for Ares F3-1333C9Q-32GAO, which doesn’t support XMP) and 2) with highest clock rates and best timings (as achieved in our overclocking tests above) set up manually in the mainboard’s BIOS.

First of all we want to run synthetic benchmarks of memory bandwidth and latency. We will use two tools for that. One is the memory test from SiSoft Sandra 2013. Using a multithreaded Stream algorithm, it is perfectly suited to modern memory controllers optimized for multi-core CPUs.

And the other is the integrated test from the ASUS MemTweakIt! utility which issues a single number indicative of the overall memory subsystem performance.

The clock rate of DDR3 SDRAM has a rather strong influence on memory subsystem performance in platforms with Ivy Bridge CPUs. For example, increasing that clock rate by 333 MHz improves the practical memory bandwidth by 10-20% and reduces latency by 5-15%. Both frequency and timings affect latency whereas bandwidth doesn’t depend much on timings.

We can also note that four 8GB memory modules turn out to be no worse than two DDR3 SDRAM modules in terms of practical memory subsystem performance. We can see this easily by using data from our previous tests. For example, the DDR3-2400 memory kit Corsair Dominator Platinum CMD16GX3M2A2400C10 (consisting of two 8GB modules and with timings of 10-12-12-31) had a bandwidth of 31.21 GB/s and a latency of 16.8 ns in Sandra 2013, which is even worse than the results of the G.Skill TridentX F3-2400C10Q-32GTX kit which has the same specs but consists of four 8GB modules.

The SiSoft Sandra 2013 and ASUS MemTweakIt! results suggest that it makes practical sense to be particular about choosing DDR3 SDRAM modules. However, these are synthetic benchmarks which are meant to measure how fast the CPU can work with system memory. Everyday applications are not so memory-sensitive, yet we want to check them out as well.

Fast overclocker memory is indeed appropriate in some situations, especially if it is priced very similarly to the regular mainstream memory kits. Memory subsystem parameters can affect the speed of gaming applications, for example.

Of course, it is the graphics subsystem that influences the frame rate the most, but the memory frequency is a factor, too. Fast memory can help you improve performance by a few percent. So again, the choice of DDR3 SDRAM is quite important, and the main parameter to consider in this case is DDR3 SDRAM frequency.

As for the sheer capacity offered by these 32GB memory kits, it doesn’t provide any benefits in terms of speed. Everyday computing tasks do not call for so much memory, but it may come in handy in special cases, for example if you process large high-resolution images or use virtual machines. On the other hand, there is no reason why you shouldn’t indulge yourself by purchasing more system memory for the future, just in case, especially as the products discussed in this review will only cost you $200-300.

 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 ]

Discussion

Comments currently: 9
Discussion started: 04/03/13 06:53:37 AM
Latest comment: 04/25/13 11:01:26 AM

View comments

Add your Comment